AtlasForge is an open-source system that generates an interactive web atlas for any gene family through automated data retrieval and analysis.

Requirements

• Python 3.12 or newer
• MAFFT for computing multiple-sequence alignment

MAFFT is optional when both atlasforge fetch and atlasforge build use --skip-clustering.

For development, the frontend requires Node.js. However, the recommended development setup uses Docker and the Dev Container CLI to automatically include all the dependencies.

Quick start

Install the complete data pipeline:

pip install 'atlasforge[pipeline]'

Then fetch, build, and serve a family:

atlasforge fetch 752 --data-dir ./my-atlas
atlasforge build --data-dir ./my-atlas
atlasforge serve --data-dir ./my-atlas

Note: atlasforge serve runs the dynamic server and generates API responses on demand. For public or high-traffic deployments, use atlasforge export with a static web server for better performance.

Here, 752 is an HGNC gene group ID. You can instead pass a text file containing one gene symbol or Ensembl gene ID per line, or a downloaded HGNC group TSV. Alternatively, you can run atlasforge fetch without any options for a fully interactive setup.

The fetch command comes with an interactive setup to guide you through processing a new gene family. Once you run all three commands, open http://localhost:8000 to see the running atlas.

The first fetch creates curation files and then pauses for user inspection. Review the family names, excluded genes, display symbols, species, and UniProt mappings in my-atlas/curation/, and then run the command printed in the terminal. Existing files are never overwritten, so you can rerun a fetch to update any missing data.

Documentation

See the documentation overview for guides to AtlasForge's files and performance.

Available Views

Genes
Gene annotation and expandable transcript details

Clustering
Protein sequence, coding sequence, and co-expression similarity

Conservation
Orthologs across species alongside their evolutionary tree

Expression
Tissue expression with anatomy diagrams linked to the heatmap

Genome Browser
Coverage and GWAS tracks around each gene, shown with its transcript models

Structure
Predicted and experimental models with membrane topology and residue-level confidence

Screenshots

See the full gallery for more examples in light and dark themes.

Structure View: Predicted and experimental models shown alongside membrane topology and confidence scores. This example uses Regions mode; a per-residue snake plot is also available.

Expression View: A tissue RNA abundance heatmap linked to anatomy diagrams, with tissues highlighted on hover.

Genome Browser View: Coverage tracks, GWAS associations, and gene models aligned on a shared genomic axis.

Fetch and build options

Use atlasforge fetch --help and atlasforge build --help for the complete command reference. The options most often used when creating a dataset are:

Option	Purpose
--curation-dir PATH	Store editable decisions outside the data directory
--gtex-version v8, v10, or v11	Choose the GTEx release; the default is v11
--gtex-file PATH	Use a local .gct.gz or Parquet expression matrix
--ensembl-release N	Pin the Ensembl release; the default is 116
--tree-source NAME	Use ensembl_compara, ncbi, timetree, or ucsc
--promote-alias-prefix PREFIX	Prefer matching aliases as display symbols
--download-predicted	Keep local copies of AlphaFold models
--download-experimental	Keep local copies of PDB structures
--gene-models-file PATH	Use a local GTF, GFF3, or BED12 instead of a GENCODE download
--browser-tracks PATH	Create the coverage curation file from an existing bigWig directory
--browser-gwas PATH	Create the GWAS curation file from an existing study directory
--local-coverage	Copy the coverage tracks into the site instead of reading them remotely
--browser-whole-genome	Keep the whole genome rather than slicing to the family's genes
--browser-bin BASES	Set the finest resolution for local coverage copies; the default is 25
--browser-max-bytes BYTES	Stop if local coverage copies would exceed this size in bytes
--no-review	Skip the pause for reviewing newly created curation files
--skip-VIEW	Omit browser, clustering, conservation, expression, or structure
--step NAME	Run only one named step; repeat to select several
--mafft PATH	Use a specific MAFFT executable during the build

Downloaded data lives in source/, personal decisions in curation/, reusable downloads in cache/, and the finished dataset in app/. See Source files if you want to edit, replace, or generate those inputs yourself.

Naming and configuration

Each setting can be supplied as an ATLASFORGE_* environment variable, in a .env file, or as a flag on atlasforge serve and atlasforge export. A flag takes priority over the environment, which takes priority over .env.

Variable	Flag	Purpose
ATLASFORGE_DATA_DIR	--data-dir	Dataset containing curation/, source/, cache/, and app/
ATLASFORGE_WEB_DIR	--web-dir	Built frontend files to serve
ATLASFORGE_APP_NAME	--app-name	Full name used by the page and web app manifest
ATLASFORGE_APP_SHORT_NAME	--app-short-name	Short name used in the app bar
ATLASFORGE_APP_DESCRIPTION	--app-description	Description shown on the Genes view and in page metadata
ATLASFORGE_FAMILY_LABEL	--family-label	Singular name used for a family member
ATLASFORGE_DOWNLOAD_PREFIX	--download-prefix	Prefix for downloaded figures, tables, and trees
ATLASFORGE_CORS_ORIGINS	--cors-origins	Comma-separated origins allowed to call the API
ATLASFORGE_HOST	--host	Address used by atlasforge serve
ATLASFORGE_PORT	--port	Port used by atlasforge serve

Naming changes take effect when the site restarts or is exported again. The dataset does not need to be rebuilt.

Deployment

While atlasforge serve is convenient for local use, a public atlas should run through Docker or a static web server for reliable restarts, HTTPS, and normal production traffic.

Docker

Build the dataset first, then point the ./data volume in docker-compose.yml at its directory and set the ATLASFORGE_* names for your atlas:

docker compose up --build

Open http://localhost:8080. The container exports the site when it starts and serves it through nginx. Put a reverse proxy in front of it when you need HTTPS; see the nginx guide or Caddy guide for more details on reverse proxying.

Static hosting

Export the whole atlas when you already have a web server or want a portable snapshot:

atlasforge export ./site --data-dir ./my-atlas
rsync -a ./site/ user@host:/var/www/atlas/

The output works with nginx, Apache, Caddy, S3, GitHub Pages, and other static hosts as long as it supports byte-range requests. It must also serve 404.html for unknown page routes while leaving missing files under /api/ as real 404 responses. You can look at deploy/nginx.conf for a complete example. Re-export after changing the dataset or the site name.

Development

The dev container is the easiest way to setup the project with Python, Node.js, MAFFT, and the project dependencies bundled. Install Docker and the Dev Container CLI, and then run:

./enter-dev.sh
./dev-zellij.sh

The frontend runs at http://localhost:3000 and sends API requests to FastAPI at http://localhost:8000. See Dev container setup for direct CLI commands and container details.

For a manual setup:

python -m venv .venv
source .venv/bin/activate
pip install -e '.[pipeline,dev]'
npm --prefix web ci
fastapi dev src/atlasforge/main.py --host 0.0.0.0 --port 8000
npm --prefix web run dev

Run the two development servers in separate terminals. Before submitting a change, run black --check src scripts, npm --prefix web run format:check, and npm --prefix web run build.

Licensing

This project is licensed under the Apache-2.0 license and is fully REUSE-compliant. Anatomogram artwork from EMBL-EBI is available under CC BY 4.0; see Anatomogram Artwork for its source, attribution, and modification details.